opensre is an open-source framework for building AI SRE agents that investigate production incidents on the user's own infrastructure. It collects and correlates alert context, logs, metrics, traces, and recent deployments to determine probable root causes and generate structured investigation reports.
Project overview
opensre targets the problem of scattered production incident evidence by combining structured incident investigation, runbook-aware reasoning, and scored synthetic RCA evaluation suites within a self-hostable framework designed for the user's own infrastructure.
Project type
AI Agent · Workflow · Evaluation & Observability
Use cases
Automation
Deployment
Refer to project documentation
License
Apache-2.0
Best for
Operations teams responsible for investigating production incidents on their own infrastructure where evidence is scattered across operational systems.
AI engineers or developers evaluating or building SRE agents who need an open reinforcement learning environment with synthetic incident simulations and scored RCA evaluation.
Key capabilities
Investigates incidents by collecting and correlating alert context, logs, metrics, traces, and recent deployments to determine probable root causes.
Generates structured investigation reports containing a probable root cause and linked supporting evidence.
Reads user runbooks and applies them during incident reasoning.
Can redact pods, clusters, and account IDs before external LLM calls and restore them in output.
Suggests next steps and can optionally execute remediation actions after an investigation.
Tails Hermes error logs, classifies incidents, and can alert through Telegram.
Runs scored synthetic root-cause-analysis suites that assess root-cause accuracy, required evidence, and adversarial red herrings.
Limitations and risks
Core workflows are usable for early exploration but are not yet fully stable, and APIs and integrations may evolve.
Sensitive identifiers can be sent to external LLM calls unless the optional reversible identifier masking capability is used. Teams should review this before adoption.
Product analytics and error telemetry are opt-out by default. Teams should review this telemetry configuration during adoption.
Getting started
Install opensre, run `opensre onboard` to complete onboarding, then run `opensre` to start the interactive shell. Note that configuration prerequisites are not fully specified in the documented flow.
Evidence and sources
GitHub project description: Build your own AI SRE agents. The open source toolkit for the AI era.
README: The open-source framework for AI SRE agents, and the training and evaluation environment they need to improve.
README: When something breaks in production, the evidence is scattered across logs, metrics, traces, runbooks, and Slack threads.
README: an open reinforcement learning environment for agentic infrastructure incident response, with end-to-end tests and synthetic incident simulations for realistic production failures
README: **One-shot investigation** — run the agent once against an alert file: